Ankit Gordhandas is the founder and CEO of Intersect Labs.
Intersect Labs offers services that enable intelligent decisions involving machine learning from spreadsheet data in 3 clicks.
Intersect Labs started with the simple idea that anyone who wants to make accurate predictions should be able to do it themselves.
In this episode, Ben and Ankit discuss...
- The convergence of no-code and data science
- Is data science just for big companies, or for startups?
- Using data as a non-technical person
- Using Intersect to predict, analyze and forecast with your data.
Ankit - Intersect Labs - Spotlight Podcast
Wed, 8/19 9:00AM • 29:46
data, predictions, companies, people, predict, machine learning model, historical data, machine learning, build, users, science, data scientists, started, code, super, communities, model, workflows, product, intersect
Ankit Gordhandas, Ben Tossell
Ben Tossell 00:00
Everybody, Stan here, founder of make, a platform teaching individuals and companies how to build custom software workflows and tools without writing code. This show explores the people behind the noco tools and the stories of folks using them to automate work and launch companies. Okay, here with us today, we have a kit from insect labs. Welcome to the show.
Ankit Gordhandas 00:22
Thanks a lot, Ben. Very nice to meet you. Thanks for having me here. Yeah, of course, why don't you give us a bit of a background about you and what intersect Labs is awesome. Yeah, happy to jump into it. Um, so let me begin with my my career background, because that'll fit nicely into the background and genesis of intersect labs. So I've been a signal processing engineer, data scientist, machine learning engineer, whatever you call it for about a decade now. went to MIT, graduated, work in a couple academic labs for a couple of years and then went to the industry and I've been been doing this for about a decade now, most recently, I was a consultant for a few small companies who wanted to get started on their machine learning journey. So I think building their first machine learning model or deploying the first machine learning model, and that's when I realized that a lot of my workflow was fairly repetitive across companies, or as you get data, I was cleaning data, I was trying out a few machine learning models. And once I found the most accurate one, I was deploying that machine learning models so my clients could make predictions. And as I started building some tools to help automate some of my workflow, and as my clients saw those tools, they wanted access to them. And so that's when I started turning that into a product. And now we have the company in a sec labs. Awesome. Yeah, that's the sort of build for yourself and then it just happens to grow into something bigger. What? When was intersec laughs When did you start that? first line of code was written just about two years ago, probably August 1 week of august 2018. We are 2019 or 2018. That's right. We officially became a company around November of that year. And then last year was February was when we released a product called our first customer.
Ben Tossell 02:22
Awesome. And then if you were in yc, which which perhaps Are you in?
That's right, we went through yc summer of 2019.
Ankit Gordhandas 02:31
Last one we did in person, right. The last one that was fully in person, winner 20 started in person but quickly had to be moved online and then summer 20, which is a current batch is completely online. Yeah, probably due to some I see.
Oh, sorry. I missed that.
Ben Tossell 02:50
So how was your experience there? It was, the
experience was fantastic. Oh, especially because it was in person but regardless, I think it was
Ankit Gordhandas 03:00
Great to be surrounded by super ambitious people who not only pushed you, but also encouraged you give you a lot of advice. And then obviously, I've never run a startup before I worked at a couple startups before. And so running a startup is a completely different ballgame, as I'm sure you know.
And it was great to hear from our regular speakers about their experiences of running startups. And then obviously, our partners are sort of our mentors. They've seen me, you know, they've helped scale thousands of companies before and so it was great to talk to them regularly and either do a quick gut check on whether you were doing things or get just very specific advice on how to proceed.
Ben Tossell 03:42
Yeah, hold on. Yeah, I mean, I completely resonate with that of building a product will be isn't going to just learn to build this stuff to this week or build this today to Oh, wait, this is a company I'm by running a company. It's a that's been a journey for me too, especially the non students realizing that I like it. was a problem. There's no need to get out of the way of the the team in some aspect. So yeah, definitely be there. Especially someone from the data recently who we brought in about, like, the fact that data scientists almost wasn't a thing. I don't know how many years ago, I might shoot myself in the foot that might take 10 1520 years. He wasn't a thing, data scientist. And then all of a sudden, it's like, you have data engineers, machine learning engineers, all of these pieces around it. I said, I think no code is like, in this barely a dog. People are doing no code stuff like they were doing data stuff. actually think Oh, see, there's more of a convergence into bringing that node stuff, in quotes to to work do you see much of that happening and you do recognize any similarities and how those, those two communities are sort of booming and and taking place?
Yeah, hundred percent by the way, you're completely accurate. Data science as a phrase was coined by DJ Patil. I think he was a data set or he was, I think, in a leadership leadership position at LinkedIn. If I'm not mistaken. And he was the first one who coined this phrase he later went on to become, I think the first chief data scientist of the United States. Oh, yeah, that's a random trivia right there.
Ben Tossell 05:26
Oh, , come up ago.
That's likely I think around 2008.
Okay, yeah. Wouldn't he do bar up and also know?
Exactly. Oh, yeah. Hundred percent. I think, just as data science became super popular, and as a concept, data science has been around for ages now, people looking at a lot of data and trying to figure out either what happened or what will happen. Oh, and then similarly, I think no code as a as a movement has been a recent phenomenon as well. And it's Great to see those two converge. When we started out, I didn't think specifically of no code. I just said, Oh, you know, someone's got to make it easier, right? Currently, you know, machine learning, or data science is limited to a few people who know both programming and have some background in statistics. Oh, and by the way, I also understand the business problem, right. And so I figured, you know, someone had to come along and make it easier for people who were just belonging to the third category, people who were really good at their business. They knew everything about their business. But yeah, I think as we've progressed on this journey, it's become abundantly clear that no code as a thing is probably here to say and more importantly, it's here to grow.
Ben Tossell 06:48
Yeah, for sure. Yeah. Obviously. I agree with that. I think it's a it's one of those things that companies, even something like webflow, which is around for a long time or has been, I mean, relatively long enough. Started we'll start off with, we're going to be a no code, or no code, why? It's just like we do this. And yeah, you don't have to do it. And then there's all movement comes around. And then people sort of pull that no code out of you and say, Yeah, you're no good tool. And we I mean, we even see it as, eventually anything that integrates with another tool, or any of the 2000 plus songs up here, like an ecosystem. Or you can do workflows with those things. So yeah, we see nocona is a huge piece that is not just here to stay, but it's been here. And it's been staying here for a while. It's just now that people are realizing the power of specifically some new tools and doing that's more what Yeah, what do you think? Why do you think now why do you think people are latching onto that now? Do you think is it computing power of what's available and like, companies like yourself, where things like data science maybe wasn't accessible enough and now, all of the And it's a click of, I'm gonna say a few buttons to do that those of complex things you may not have been able to do before.
Yeah, I think that's a good question. So specifically, data related stuff. Um, I think the first time people started taking collecting data seriously was probably around the time Google and Facebook were becoming big. Right. So that was the late 2000s, when people really started understanding the value of data. And then in the following five years, you saw a lot of businesses just saying, Oh, yeah, we're gonna collect data, and we'll figure out what to do with it later. Yeah. And then it during that time, a lot of people have started getting employed as data scientists, as these experts who could take whatever data that businesses has been collecting into and turn them into some insights or predictions about the future of and I think, more recently, what what I've personally seen happen is that
Ankit Gordhandas 09:00
A lot of businesses started collecting data probably around like five years ago. That's when a lot of like small and medium sized businesses started taking collecting data or more seriously. And so we're at a point where compute power has become crazy powerful, and just crazy cheap, right? It does pennies to train a machine learning model. Now, we're also at a point where businesses have been collecting good data for 2345 years. All right. And so I think this is a magic moment where a lot of business people want to start using that data and putting it into action, right, making good use out of it all. At the same time. You know, like you said, that flow has been around for a few years, I think, almost half a decade now, if I'm not mistaken. Oh, and two people are people are starting to see Oh, yeah. You know, previously, website development was limited to engineers or designers. Now I can build a website, why can't I turn my data into prediction? So I can't
slice and dice my data on myself without having to rely on someone specialized on it. That's why I think this is the magic moment for chocolate plus data.
Ben Tossell 10:08
Yeah. And you said about loans. It's the big companies that are taking the when they took notice or made it almost popular by saying yeah, we're using data to do all this fancy stuff. I mean, yesterday, I think it's called money come from AWS. Okay. Now, Google may be what they acquired was actually before the Microsoft Power Apps has been around since 2015. Or something is actually now more popular. Be also said that there was like these experts who started like getting hired by companies or came into the into the crowd and showed off what they could do how, what something comes companies, Muslims, what's up the communities as those were crucial. And helping that happen showing off. data isn't data science isn't just this thing you hear about like, this is how it works is how powerful is this one? I mean, I know of cable which, right? I know I've known for years. And actually, I think Google more than right. And there was a second video. And there's the idea of it years and years ago. And I think that, yeah, it's just cool to see that. So I wonder if there's anything specific that you've seen something. On the other side, you see dribble, and then that's such a good option for designers, GitHub credentialing for all types of developers. I wonder if you've seen anything else in the data science world that, you know, might be interesting for me to take a look at because I just interested in all these different communities and sort of credentialing of that.
Yeah, the question I think kaggle is by far, the most powerful community for anything data related, all right. It's also great because Until kaggle came along, if you wanted to learn data science stuff or practice or like data manipulation skills, you are limited to whatever data you could control yourself. Right? And then came along kaggle. And suddenly they had these challenges that these leaderboard competitions, where you could just get access to really cool data, right? It started off with simple tips, tabular data. And now today, you can get images, videos, audios everything, right. And so I think kaggle is is by far the leader. In terms of communities, there is also a bunch of communities on LinkedIn and Facebook, which, you know, sort of support each other. Or there's a few slack groups out there, where as a data scientist, you can volunteer your time or to help like to cool analysis for public causes and stuff like that. So there's these little pockets of niche communities and then something else That's, that's super fun in terms of like how data science became more visible, I think is the story of Facebook's growth team. I think it was it was a mark of social capital who assembled a growth team at Facebook with like super data savvy people. And they slice and dice like all kinds of data. And they came out with this thing where they realized if someone had a D seven friends on Facebook, they were extremely low, likely extremely low likelihood of churning as a user. And so from that point on Facebook have directed all their energy in their product to make sure that as soon as you sign up, you sign you added at least seven hands right and that was magical. And then all those other story that target came out where they, they, I think they the data scientist, or the data science team, analyze purchases, cheese and Started recommending products to buy in little flyers that went out by mail to people's houses, and they were personalized. Oh, and there's one person based on the purchases tree. And they were able to figure out that this person was pregnant or likely pregnant. And so they started recommending pregnancy, maternity wear and stuff like that. Um, but she hadn't declared to her family that she was pregnant.
Ben Tossell 14:27
I think so this exact story in a podcast earlier today of I think it was from my first million with sandpile. And Sean, and they talked about, there's something about in terms of data privacy, and yeah, it's like, yeah, it's this young girl, I think was younger when exactly stuff getting delivered to her father or something.
Yeah, and and obviously, that was the first time you know, people started raising concerns about privacy right yourself, all right, because there's obviously ethical questions around what you can do with data. But that was also one of the first things As people started realizing the power all that data could have, right? It's just, it's just a bunch of numbers at the end of the day, but together, it can be it can like do really powerful things.
Ben Tossell 15:11
Yeah. I mean, it could be make or break, right. So tell me a bit more about intersect labs. And what do you do there? And how, how companies users are using right now?
Yeah, hundred percent. So yeah, the way the product works is super simple. Like I mentioned earlier, the philosophy is that anyone who has historical data can start turning their historical data into predictions. And that's powerful, because until now, that process has been limited to a few very highly few specialized individuals, data scientists who are good at programming, math and statistics and have the domain expertise. Oh, and until now you had business analysts or data analysts were really good at the business problem. They were really good at collecting all the historical data, or, but unfortunately, they were limited to in their analysis to explain what happened in the past rather than predicting what will happen in the future. So what happens now is that the same data analysts have and I lose that title loosely, it's, it's basically anyone who analyzes data, oh, oh, they can bring their data in a spreadsheet format, oh, upload it to our platform and select which column they want to be able to predict. And once they do that our platform takes over or it'll first clean the data, which is to say that if you have missing values, or if you have strings in your data set, or if you have outliers, or things that are typically bad for machine learning, our platform will take care of all of that automatically. In lens like a few machine learning algorithms from from the repository that we have, usually between three and five, to train, it will train all of those algorithms multiple times and as it's doing so it's also assessing the accuracy of and once it identifies a super accurate model. model is deployed. So as a user, you can start making predictions right away. And you can do that one of two ways. You can either use our GUI, just upload a new spreadsheet, get a spreadsheet and return with predictions, or we have an API. So if you want to build a machine learning model into your stack, all you do is send as a JSON or CSV, you and your data and as a JSON, you would get back all the predictions that you need. Oh, and that's super powerful, because a number of use cases are a subsidiary of Philips health, for example, uses us to predict which patients will likely not show their doctor's appointments. Right? Then we have a pretty, pretty big company who's one of the like, if I tell you to name all the companies that have the best data science talent, this would be the top three. They actually use. The business development team uses us because the business development team did not have access to the data scientists and so they use us to have predict or prioritize, which leads to talk to based on the usage patterns that the users have on the platform. And then you have a snack box, catering company or migraine. I think they're now rebranded as carton leaves us for everything from demand planning, forecasting inventory, um, logistics and supply chain management, predicting their warehouse costs. We have a company, a small company in Massachusetts that helps people apply for disability insurance and they use us to sort of predict which patient which which other users will drop off and proactively reach out to them most of the times to make sure that those users don't drop off. Um, so yeah, there's a bunch of different use case ecommerce companies in us to predict their sales in the coming months. Figure out price elasticity. So I think, you know, hey, if I do a 10% off promotion next month, what impact will it have on revenue? So there's a whole lot of different use cases that we have. The one common thing across every single user is that they have historical data. They want to turn that historical data into predictions.
Ben Tossell 19:11
Awesome. Yes. It is lignite across the whole operations of the business, anything really it gets everything remembers a certain point you can even do. That's awesome. But um, what's so what's Where's where is the company now? And what's the what's next for 32nd rounds?
Yeah, the question. So we're a small team. So a team of four individuals. And super close knit even though we're all remote. Actually. I was quite like a nomad. Ah, yeah, um, we are in terms of what's next. Actually, most recently, last week, we released a new product called pasture. And the idea for that product is early on as customers wanted to use their data for machine learning. We were doing a lot of data restructuring or data wrangling for about using custom code. And so we started building some internal tools to help us do that easier. And again, when our customers saw some of those tools, they wanted access to them. So we turned that into a user facing product. We call it pasture because it pasteurized your data quite like pasteurization. Okay. And yes, that product has been released last week as a standalone tool, you could use it or right before you use machine learning, or you can use it just to wrangle your data. And we've had some interesting use cases come out there as well. Or you have a company or actually a fellow batchmate from yc legacy, or they send out these giant service to their users to better help them and they use pasture to run decision trees on their survey. So think things like or if you run at least five miles a week, but you know don't eat enough like proteins and don't take supplements. Okay? We recommend that you start taking some supplements or if you don't walk at all, but he like five servings of bread in a day that he asked you to I started walking while he's not bad, that kind of stuff.
Ben Tossell 21:03
Yeah, or like, you know, a VC firm is using us to do their due diligence analysis on on, like, incoming deals. So yeah, that's that's where we are right now I think moving forward. In general, we try to be as customer centric as possible, or so if a customer says, Hey, we need this, or if we observe a customer needing something, we try to build it into our, into our platform. And so I think the upcoming thing is building workflows. Because at the end of the day, data science is last about just training a model or painting or data, it's that end to end thing where you have your data, you clean it, you train the model, and then you productionize it and production ization can often mean you know, every Saturday morning, you want your data in your database to be pulled predictions on on predictions dumped into different tables in the database, right? All until we're is starting to build. Again, we have some internal tools to support that we're starting to build that into user facing product where users can come in and start defining some of the workflows.
Ben Tossell 22:11
Awesome. It seems like this type of I mean, to me, generally, it seems like data science will prediction predicting things. There's almost when you've got things almost figured out, or you've you're, you've figured something that will seems the it sort of lends itself to already established companies that have revenues and customers have certain things. Is that right? is there is there space for with no, no good movement in general. We see a lot of people who identify themselves as a entrepreneurial founder, which may not be already that but they work somewhere else. But they're just using that as the power of discovering no code and how they can build things. And then is it wears too ready to start thinking about using data in what you're doing much? To me, it seems as if so many people start companies and do mdps. And they celebrate when they get to $100 monthly recurring revenue, which is great if it never sold anything, but it could be opportunity that it could have started using stuff before isn't all qualitative, which is often the case that early stage.
Yeah, that's a good question. Something you said early on. And this question struck me I think you said, Do you like predicting stuff has, you know, seems like it's better suited for big corporations who have who have stuff figured out. And that's been true for the most part until now, mostly, because until now, you have to rely on data scientists. And data scientists are expensive because they're super talented, right? think someone who is really good at math and programming. Often a PhD is you know, you obviously want to pay them a decent salary. And so which is why predictions have been limited to some of the bigger companies. But smaller companies often have a lot of data. All right, enough data to just start making turning that that data into predictions. And that's where all the come in, right? I'm sure right? Someone who's just built an MVP has $100 in monthly recurring revenue, probably doesn't have enough data to start building machine learning models. Oh, but a company that's been around for you know, maybe a decade or so makes easily like hundred k a month. They might not have the ROI. justified justification to hire data scientists, but they would definitely be able to afford or our product and and start, you know, really making massive either savings or revenue out of using us.
Ben Tossell 24:51
Awesome. Yeah. So yes, you would need a set number of lines of data before it's really worth like Was I guess it can even train the models and fold out anything useful. Right.
Exactly. And and if you had an FAQ page, which is coming shortly, the the number one question would be how much data is enough? And? And unfortunately, the answer to that is it depends, right? It depends on the kind of accuracy you're looking for. But yeah, you're right. I think, if you have like 100 rows of data, that's never going to be enough to build a machine learning model out of it. Um, but the good thing, the good news is that we always, you know, anyone we talk to, if they're interested in figuring out if they have enough data, we just tell them, hey, send us your data, or let's hop on a live demo. And we can train a machine learning model live. And then let's figure out right if it's accurate enough or not. If it's not accurate, if you don't like it, then pay keep collecting data, come back to us, maybe in a quarter, maybe in two quarters, and you'll be there and we often give advice as to what other data that they can collect is. Um, yeah, the other thing is over the new product that we released last week, Patrick You don't need a lot of historical data or any historical data, right? Because it's mostly just been doing either decision trees or some data wrangling workflows. Which is how, which is why legacy started to use us legacy. They're fairly new company. Like I said, there are batchmates and yc. Also not enough historical data, but they still wanted to run decision trees. And they came up with a simple heuristics based approach, designed by an epidemiologist. And they were able to use pasture for a lot less money than a machine learning machine learning product. And, you know, now that they're collecting data, hopefully in three to six months time, they'll have enough data and we transition
Ben Tossell 26:42
them over to the machine learning side, also, is them. I see with all the data that you're using, are people using you for predictions and saying they're expecting an accuracy? Do you ever worry about or assume you probably have had Once or twice where people are saying, this didn't quite come out as you as your as you all like saying your model and as your model predicted, how'd you how'd you get around and things like that?
Yeah, fortunately, that hasn't come up too much. More often than not, people are pretty surprised that it when they deploy the model in the real world, it comes out the predictions are natural, or with their expectations. But you know, having said that, at the end of the day, it's a model right? Especially when it comes to predicting human behavior sort of data that is always going to be inaccuracy is right humans are irrational creatures at best. And yeah, yeah. So at the end of the day, we you know, we talk to them often, they often understand that things can get inaccurate, but in general in our global picture, I have a picture view on average things are more accurate with machine learning model than without and then Obviously, we encourage them to retrain their models every few months to make sure that all the new data that they've collected is factored in.
Ben Tossell 28:08
Yeah. Okay, awesome. Well, yeah, I mean, I really appreciate you jumping on and talking about all this because we're really interested in and not something not the sort of rabbit hole easy dive down. So yeah, it's been great having you on. So when you tell the folks listening, where they can find you and where they can find these routes.
Yeah, for sure. Thanks a lot for having me here. But if you're interested in getting started with machine learning, or just talking to us to explore whether you can get started with machine learning, just hop on over to intersect labs.io. You'll see all the information you need. They'll also be a button to book a call, most likely with me or one of our teammates was the other three teammates. We're all we've all been doing this for a long time. So we'll be happy to take care of you. And then if you want to start using our newest product faster, you don't even need to talk to us. Just hop on over to intersect labs.io forward slash pasture. p s t u r. And I think we have a link to it on maker pad as well. Oh, and you can just get started. We have a two week free trial. And then if you like it, I'll keep using it.
Ben Tossell 29:16
Awesome. Well, thanks. Just show off some of the stuff and again, get that no code and data science stuff together and show that off and help each other soon. Awesome. Thanks a lot, man. Cheers. Thanks so much for listening. You can find us online at maker pad.co or on Twitter at make Pat. We'd love to hear if you enjoyed this episode and what we should do next.